AI website builders are real and they work: you describe your business in a conversation and, within minutes, you have a page with text, images, and a live address. Our agency's own site was built this way. The part nobody shows in the demo is what comes after — and that's what decides whether the site brings in customers or just sits there.
The three types of AI website builders
The name is the same, but these are quite different tools. Knowing which one you're using changes what you can expect:
- A chat assistant that publishes a page. This is the case with ChatGPT Sites: you ask, it generates and hosts. OpenAI itself positions the tool for dashboards, prototypes, and internal pages, and the documentation doesn't address how these pages show up on Google.
- An AI website builder inside a hosting platform. Wix and Hostinger are examples: the AI puts together a first draft from a template, and you adjust it in a drag-and-drop editor. Hosting and domain come in the same package.
- A tool that writes the site's actual code. Lovable, which we use, is this type: the AI generates real code, and you can take it anywhere. It offers more freedom — and requires someone to maintain the site after the conversation ends.
What it really costs
The sticker price rarely matches what you pay the following year. Two examples checked on the official sites in October 2026:
- Hostinger (Brazil): the entry-level website builder plan costs R$ 10.99 per month, but only on a 48-month contract. On renewal, the same plan jumps to R$ 38.99 per month. The domain is free only for the first year.
- Lovable: charges by credits, and every change request uses some up. The free plan gives 5 credits a day, up to 30 a month — enough to test things out, but short for a full site with revisions.
Before signing up, work out the renewal cost and how much each change will run you. That's where the gap between "cheap" and "cheap for the first month" lives.
What happened with our site
The AI quickly delivered what it promises: structure, pages, a coherent look, and starter copy. That saved weeks. The problems showed up afterward, all in things you can't see just by looking at the screen:
- The blog was hidden from Google. The article list pages told Google they were a copy of the homepage. Result: of 48 articles published over four weeks, only 6 were in Google's index. After the fix, 17 got indexed in four days, and new articles started getting indexed within a day or two.
- The same site answered at four different addresses — with and without "www," with and without "https." Google even showed the wrong version instead of the right one. Today they all lead to a single address.
- The page was too heavy on mobile. Our website-creation page downloaded 4.8 MB of images on a phone. Converted to a lighter format, that dropped to 1.6 MB, with no visible difference.
- The contact form was slow to notify us. A customer's message sat in storage until a routine forwarded it by email — and that routine only ran once a day. A lead could wait almost 24 hours. Now it runs every 15 minutes.
None of these issues show up in the preview the AI displays. All of them affect whether customers find the site and whether their message reaches you.
When AI alone gets the job done
- An event, campaign, or launch page with an expiration date.
- A prototype to pitch an idea to a partner or investor.
- An internal team page that doesn't need to be found on Google.
- A first site for someone still testing a business who'll overhaul everything in six months.
When it's worth having someone alongside
- The site needs to bring in customers through Google — which is to say, almost every business site.
- You're changing addresses and already have indexed pages that can't be lost.
- The site needs to exist in more than one language, each with its own address.
- There's a form, booking system, or order process that can't fail without you knowing.
It isn't a choice between AI and people. The cheapest path is usually both: the AI builds the draft in hours, and someone checks what it doesn't check. That's how we work on website creation.
What each type solves — and what it leaves out
Chat assistant (like ChatGPT Sites). It solves speed: in minutes the page exists and has an address. It leaves out almost everything a business needs in the long run. OpenAI's documentation does not cover how these pages show up on Google, the tool only works on paid plans, and there is no staging area to review before publishing — everything goes live right away.
AI builder inside a hosting platform (Wix, Hostinger). It solves maintenance: hosting, security and domain come in the package, and the drag-and-drop editor lets the owner change text and photos without calling anyone. It leaves out the freedom to leave: the site is built inside the platform and, if you ever want to switch providers, you usually rebuild it from scratch. The renewal price also weighs in, as shown above.
Tool that writes the code (Lovable). It solves freedom: the code is yours and can go to any server. It leaves out maintenance: someone has to host it, publish changes, check what Google reads and fix things when they break. That is the path we chose for our own site — which is why the four problems above landed on us to solve, not on the tool.
The most common mistakes when publishing on your own
- Keeping the generic text the AI wrote. “We are a benchmark in quality and service” fits any company in the world. Replace it with what only you do: the neighborhood you serve, your most requested service, your real delivery time.
- Keeping stock photos. Customers spot a generic photo from a mile away. An honest photo of your storefront, team or product is worth more than the prettiest stock image.
- Forgetting each page's title and description. That is what shows up in Google's results. If every page has the same title, Google doesn't know which one to show.
- Publishing without testing on a phone. The tool's preview is usually on a computer. Most customers will open it on their phone.
- Registering the domain in someone else's name. Whoever registers it owns the address. If it was your nephew or the tool, your site depends on their goodwill.
Checklist for after the AI is done
- Does the site show up on Google? Register it with Google Search Console, which is free, and see how many pages are indexed.
- Is there just one address? Test with and without "www": both should lead to the same place.
- Does it load fast on mobile? Test on 4G, not the office Wi-Fi.
- Does the form actually reach you? Send a test message and time how long it takes to arrive.
- Is the domain in your name? If it's registered under the tool's name or the developer's, you don't own the address.
Sources
- Pricing pages for Hostinger (Brazil) and Lovable, checked in October 2026.
- ChatGPT Sites documentation, OpenAI (public beta since July 2026).
- Data from the agency's own site: Google Search Console and page-weight measurements, August and September 2026.
Frequently asked questions
Is there AI that builds websites on its own?
Yes. Tools like ChatGPT Sites, Wix's and Hostinger's AI builders, and Lovable can put together and publish a page from a description in just a few minutes. What they don't do on their own is check whether the site shows up on Google, loads fast on mobile, and whether your contact form actually reaches you.
What's the difference between ChatGPT Sites, Wix, and Lovable?
ChatGPT Sites generates and hosts pages right inside ChatGPT, and OpenAI positions it for dashboards, prototypes, and internal pages. Wix and Hostinger use AI to put together a draft in an editor with hosting included. Lovable writes actual site code that you can take elsewhere, but it still needs someone to maintain it afterward.
How much does it cost to build a website with AI?
There are free plans to try it out and paid monthly plans. Watch out for renewal pricing: on Hostinger Brazil, the entry-level plan runs R$10.99 a month on a 48-month contract and renews at R$38.99. On Lovable, every change request uses up credits, and the free plan gives you 5 a day, up to 30 a month.
Does a website built by AI show up on Google?
It can, but it's not automatic. On our own agency's site, built with AI, only 6 of 48 articles were showing up on Google because of a technical detail in the blog listing; after we fixed it, that jumped to 17 in four days. Register your site with Google Search Console and check how many pages are actually indexed.
When is it worth hiring someone instead of relying only on AI?
When the site needs to bring in customers through Google, when you're changing domains and can't afford to lose pages that are already indexed, when you need more than one language, or when there are forms and orders that can't fail without you knowing. For an event page, a prototype, or testing a business idea, AI on its own is usually enough.
What should you check after AI builds the site?
Whether the site shows up in Google Search Console, whether there's just one canonical address (with and without www leading to the same place), whether it loads fast on mobile over 4G, whether the form message actually reaches you, and whether the domain is registered in your name.


